Relationship of Road Lane Width to Safety for Urban and Suburban Arterials Phd. Alma AFEZOLLI Polytechnic University of Tirana Construction and Infrastructure Department of Civil Engineering Faculty Tirana, Albania krasniqi1@yahoo.com Phd. Elfrida SHEHU Polytechnic University of Tirana Construction and Infrastructure Department of Civil Engineering Faculty Tirana, Albania elfridaal@yahoo.com Abstract This paper investigates the relationship between lane width and safety for roadway segments on urban and suburban arterials. The paper found no general indication that the use of lanes narrower than 3.6 m on urban and suburban arterials increases crash frequencies. This finding suggests that geometric design policies should provide substantial flexibility for use of lane widths narrower than 3.6 m. Inconsistent results were found, which suggested increased crash frequencies with narrower lanes in three specific design situations. Narrower lanes should be used cautiously in these three specific situations unless local experience indicates otherwise. (Abstract) Keywords lane, width, safety, arterial, accident, frequency. I. INTRODUCTION This paper addresses the relationship of lane width to safety for urban and suburban arterials. A cross-sectional safety analysis approach was used, because suitable sites to conduct a before-after observational study were not available. Lane width for midblock segments has been considered. A full report of the results of this research has been prepared. II. CURRENT STATE OF KNOWLEDGE ON SAFETY EFFECTS OF LANE WIDTHS The conventional wisdom of most infrastructures engineers is that use of narrower lanes in the design of a roadway, will result in more crashes, if other design characteristics of the roadway remain unchanged. This has been demonstrated for lane widths on rural two-lane highways [2], but there is no definitive research on the safety effect of lane widths for urban and suburban arterials. The use of narrower through-traffic lanes may have advantages in some situations on arterials, because this may reduce pedestrian crossing distances and may provide space for additional through lanes, auxiliary lanes, bicycle lanes, buffer areas between travel lanes and sidewalks, and placement of roadside hardware. However, the use of narrower lanes has been accompanied by concerns that reducing lane widths could increase crash frequencies. A number of past studies have been conducted to determine the traffic safety effects of lane width, but the results of these studies are varied. Separate models were developed for offthe-road and on-the-road accidents. Hauer concluded that for off-the-road accidents, if accident frequency is influenced by lane width, it is not discernable. [1] For on-the-road accidents, lane width was found to be associated with property-damage only (PDO) accidents, but not injury accidents. In the PDO model, Hauer notes that wider lanes are associated with higher accident frequencies (not lower accident frequencies.) However, Hauer notes that the relationship is weak, and lane width is only included in the model, because of the traditional interest in this variable. Hadi (5) developed negative binomial regression models to estimate the safety of various cross-sectional elements and found that increasing lane widths up to 3.7 m and 4.0 m would be expected to decrease accident rates for urban two-lane roadways and urban four-lane undivided roadways, respectively. While many countermeasures have been identified to reduce crossing distance at intersections for pedestrians, no studies have documented the quantitative effect of lane width on pedestrian or bicycle safety. III. CURRENT GEOMETRIC DESIGN POLICIES FOR LANE WIDTH ON URBAN AND SUBURBAN ARTERIALS Highway design policies for arterial roadways indicate a preference for the use of 3.6 m lane widths, but also indicate flexibility for use of narrower lanes where 3.6 m lanes are infeasible or impractical [6]. The geometric design practices related to lane width must consider the needs of motor vehicle, pedestrian, and bicycle traffic. The American Association of State Highway and Transportation Officials (AASHTO) offers guidelines on the selection of appropriate lane widths on urban and suburban arterials considering primarily the needs of motor vehicle traffic, where lane widths from 3.0 to 3.6 m are addressed along with specific circumstances, for which each width should be considered. The purpose of this paper is to investigate, if this conventional wisdom is correct, determining whether and how lane widths affects safety and identify situations in which design flexibility to use narrower lanes should or should not be utilized. 119
IV. SAFETY EVALUATION OF LANE WIDTHS ON ARTERIAL MIDBLOCK SEGMENTS A. Available Data Base A database is used in this paper to examine the effects of roadway features, including lane width, on safety for arterials. The objective is to develop a prediction methodology for urban and suburban arterials for application in the forthcoming Highway Safety Manual. This database is also suitable for investigation of the relationship between lane width and safety and has been used for that purpose in the current research [7]. The database includes site characteristics, traffic volume, and crash data for arterial roadway segments in Tirana and Durres cities of Albania. The roadway segments in Tirana include roadways in both urban and suburban communities. The roadway segments in Durres include some urban communities, but most of the area would be considered suburban. The database for both areas includes a mixture of arterials under state and local jurisdiction. The available data include five arterial roadway types: two-lane undivided arterials (2U) three-lane arterials including a center TWLTL (3T) four-lane undivided arterials (4U) four-lane divided arterials (4D) five-lane arterials including a center TWLTL (5T) Table 1 presents a summary of the number of sites, for which site characteristics including lane width, traffic volume, and crash data are available in each city. Each site consists of one block (e.g. the arterial roadway from one public road intersection to the next). The blocks range from 0.06 to 2.28 km with an average block length of 0.21 km. Table 2 shows comparable data for the total lengths of sites in each table. The database includes 408 km of urban and suburban arterials (130 km in Durres and 278 km in Tirana). Table 1 Number of roadway segment analysis sites by roadway type and lane width Roadway Type TIRANA CITY Number of sites by lane width (m) 9 10 11 12 13+ Total 2U 2 20 20 162 176 380 3T 16 5 73 35 129 4U 19 147 121 91 62 440 4D 2 44 61 71 178 5T 8 18 26 Subtotal 21 185 198 405 344 1153 DURRES CITY 2U 61 82 229 148 70 590 3T 31 125 12 32 15 115 4U 104 81 157 29 471 4D 12 10 69 33 16 140 5T 17 357 114 74 562 Subtotal 208 315 824 356 175 1878 TOTAL 229 500 1022 761 519 3031 Table 2 Total length of roadway segment analysis sites by roadway type and lane width Roadway Type TIRANA CITY Total length of sites (km) by lane width (m) 9 10 11 12 13+ Total 2U 0.19 2.86 3.34 21.02 24.88 52.29 3T 4U 4D 5T Subtotal DURRES CITY 2U 3T 4U 4D 5T Subtotal 1.27 0.65 11.96 4.14 18.02 0.5 5.79 11.29 13.81 31.39 1.43 13.74 14.15 12.77 5.4 47.49 1.57 2.71 1.62 18.37 25.5 59.75 48.23 153.47 7.5 11.45 33.9 23.81 11.4 88.06 4.11 2.52 1.8 5.36 1.96 15.48 7.76 14.28 14.17 4.07 40.28 0.78 9.59 12.01 4.42 2.83 29.63 1.16 50.78 19.3 10.27 81.51 20.15 39 112.66 55.96 26.46 254.96 TOTAL 21.77 57.37 138.16 115.71 74.69 408.43 The lane widths at these sites were measured in the field. The lane width shown in the table represents the average lane width across all through travel lanes. Sites for which measured lane widths were not available have been omitted from Tables 1 and 2 and from the subsequent analyses. Lane width category Table 3 Range of lane width (m) 9 2.85 or less 10 2.85-3.15 11 3.15-3.45 12 3.45-3.75 13+ 3.75 or more Crash data were obtained for all of the sites shown in Table 1 for a five-year period: 2008 to 2013 in Tirana and 2007 to 2012 in Durres. The crash data included 4,786 crashes in Tirana and 3.054 crashes in Durres. The analysis performed in this study included all multiple- and single-vehicle accidents, except pedestrian and bicycle collisions. B. Analysis Approach Used The best method to determine the effect of a roadway geometric feature on safety is through a well- designed beforeafter evaluation. While the use of a before-after evaluation would be the preferred approach to determining the effect of lane width on safety, a before-after evaluation was not feasible in this paper, because highway agencies seldom change the lane width of a roadway, without making other changes that would confound the results of any before-after evaluation conducted. Two approaches to cross-sectional analysis to examine the effect of lane width have been applied in this paper. [4] Each approach was applied separately to data from each roadway type. In the first approach, only three variables 120
were considered: average daily traffic (ADT) volume, roadway segment length, and lane width. In the second approach, a broader set of site characteristics were considered in addition to ADT, segment length, and lane width. The first approach began by developing an ADT-only negative binomial regression model in the form: where: N = exp (a + bln(adt) + ln(l)) (1) N = predicted number of crashes per year of a particular crash type ADT = average daily traffic volume (vehicles/day) on the roadway segment L = length of roadway segment (m) a, b = regression coefficients In addition to the ADT term, the ADT-only models for roadway segments also included the roadway segment length as a factor representing exposure. [5] Then, models were developed in the same form as equation (1), but with a set of variables added to represent the effect of lane width: where: N=exp(a+bln(ADT) +ln(l) +c 9 LW 9 +c 10 LW 10 +c 11 LW 11 c 12 W 12 +c 13 +LW 13 +) (2) LW 9 = indicator variable (= 1 if lane width of roadway segment = 2.7 m; = 0 if not) LW 10 = indicator variable (= 1 if lane width of roadway segment = 3 m; = 0 if not) LW 11 = indicator variable (= 1 if lane width of roadway segment = 3.3 m; = 0 if not) LW 12 = indicator variable (= 1 if lane width of roadway segment = 3.6 m; = 0 if not) LW 13 + = indicator variable (= 1 if lane width of roadway segment = 3.9 m or more; = 0 if not) c 9,..., c 13 + = regression coefficients Lane width was treated as a categorical variable in this modeling approach, rather than as a continuous variable, because there was no reason to presume a linear or log linear relationship between lane width and safety. Treatment as a categorical variable provides an opportunity for unusual or unexpected relationships between lane width and safety to be identified. Lane width effects were included in models in the form show in equation (2), only if the effect of lane width was found to be statistically significant. [2] The second approach began with the best models developed considering variables, other than lane width. These models were typically in the form: where: N=exp(a+blnADT+lnL+dSW+eOSP+fRHR) (3) SW = shoulder width (m) OSP = on-street parking indicator (= 0 if curb parking is present on either side of street; = 1 if not present) RHR = roadside hazard rating for roadway segment (1 to 7 scale) d, e, f = regression coefficients The shoulder width, on-street parking, and roadside hazard rating variables were included, only if their coefficients were statistically significant. The roadside hazard rating used was a rating on a scale from 1 (best roadside) to 7 (poorest roadside) developed in research by Zegeer [8]. To this best model, in the form shown in equation (3), the current research then added the same lane width effects that were considered in equation (2): N=exp(a+bln(ADT)+ln(L)+d SW+eOSP+fRHR+ c 9 LW 9 +c 10 LW 10 +c 11 LW11+c 12 LW 12 + c 13 +LW 13 +) (4) Lane width was added to equation (4) only if its effect was found to be statistically significant. Nine dependent variables [represented by N in equations (1) through (4)] were considered: All crashes Fatal-and-injury crashes Property-damage-only crashes All multiple-vehicle crashes (nondriveway-related) Fatal-and-injury multiple-vehicle crashes nondriveway-related) Property-damage-only multiple-vehicle crashes All single-vehicle crashes Fatal-and-injury single-vehicle crashes Property-damage-only single-vehicle crashes Both analysis approaches were applied to: 2 model forms either equation (1) and (2) or equations (3) and (4) 9 dependent variables 5 roadway types Thus, a total of 90 regression models were developed for each analysis approach. The results of these modeling approaches are presented below. [4] V. ANALYSIS OF RESULTS All of the 45 models of Tirana roadway segment crashes using the ADT-only model in the form shown in equation (1) were statistically significant with R LR 2 ranging from 0.08 to 0.45. Lane width variables were added to create models in the form of equation (2). Table 4 shows the coefficients of the lane width variables [c 9, c 10, c 11, c 12, and c 13 + in equation (2)]. The coefficients are 121
all expressed through comparison to a nominal lane width of 3.6 m (i.e., the value of coefficient c12 is always zero). Positive coefficients indicate that roadways with the corresponding lane width would be expected to have higher crash frequencies than roadways with 3.6 m lanes. Negative coefficients indicate that roadways with the corresponding lane width would be expected to have lower crash frequencies than roadways with 3.6 m lanes. The values of the coefficients must be interpreted in accordance with equation (2). The actual effect of lane width on safety is determined by taking the exponential function of the coefficient [e.g., exp(c 10 )]. The final two columns in Table 4 indicate the results of comparisons of the coefficients for different lane widths. The next-to-last column indicates a comparison of the lane width effects for 2.7 and 3.0 m lanes to those for 3.3 and 3.6 m lanes. The comments in the last column of Table 4 (and subsequent tables) are interpreted as follows: Table 4 Negative binomial regression models with ADT and lane width for roadway segments in Tirana Decrease means that 3.3 to 3.6 m lanes have lower crash frequencies than 2.7 to 3.0 m lanes. This is consistent with the conventional wisdom that wider lanes result in lower crash frequencies. Increase means that 3.3 to 3.6 m lanes have higher crash frequencies than 2.7 to 3.0 m lanes. This is opposite to the conventional wisdom. No change means that the crash frequencies for 3.3 to 3.6 m lanes are so close to those for 2.7 and 3.0 m lanes that there is little practical engineering difference between these values. Inconsistent means that the crash frequencies for 2.7 and 3.0 m lanes fall between those for 3.3 and 3.6 m. Table 4 shows that when the lane width variable was added to the 45 statistically significant ADT-only models: in 17 cases, statistically significant models involving both ADT and lane width were found. in 22 cases, no model was found (i.e., the modeling algorithm did not converge). This indicates that the addition of the lane width interfered with the relationship between safety and ADT that had already been determined. in 6 cases, statistically significant models were found but the value of R 2 LR was so low (below 0.10) that the model has little predictive ability. In these cases, the 2 ADT-only model had R LR above 0.10, so the predictive ability of the model including lane width was less than the ADT-only model. [5] In the 28 cases for which no model was found or a model with R LR 2 below 0.10 was found, there is no indication of a strong relationship between lane width and safety. In the 17 cases where both ADT and lane width had a statistically significant effect, there were only nine cases in which the effect for lane width in the range from 2.7 to 3.6 m was in the direction expected by the conventional wisdom (i.e., decreasing crash frequency for wider lanes). These nine cases included all of the dependent variables considered for one particular roadway type- four-lane undivided roadways. In general, for four-lane undivided roadways on Tirana arterials, roadways with lane widths of 3.0 m or less were found to have higher crash frequencies than comparable roadways with 3.3 or 3.6 m lanes. There is no indication in the Tirana data of a consistent relationship between safety and lane width for any other roadway type. It should be noted that the Tirana data contain relatively few sites with 2.7 m lanes. Therefore, the finding noted above generally indicates that four-lane undivided arterials in Tirana with 3.0 m lanes tend to experience more crashes than those with 3.3 and 3.6 m lanes. [8] Table 5 presents comparable results to Table 4 for arterial roadway segments in Durres city. The results are comparable to the Tirana results in that there were only a limited number of statistically significant models incorporating both ADT and lane width. Specifically, out of the 45 cases for which statistically significant ADT-only models were found: 122
in 25 cases, statistically significant models involving both ADT and lane width were found. in 4 cases, statistically significant models were found but the value of 0.10. R LR 2 was below 0.10 in 16 cases, no model was found (i.e., the modeling algorithm did not converge). The Durres data do not show a lane width effect for fourlane undivided roadways similar to that found in Tirana. Fourlane undivided roadways with 3.0 m lanes in Durres generally had crash frequencies comparable to roadways with 3.3 and 3.6 m lanes. The only pattern noted was that for four-lane divided arterials in Durres, roadways with 2.7 m lanes tend to have higher crash frequencies than roadways with 3.0 m lanes. [3] There is no indication in the Durres data of elevated crash frequencies for 3.0 m lanes on four-lane undivided roadways as found for Tirana. There is an indication in the Durres data that higher crash frequencies may be found for 2.7 than for 3.0 m lanes on four-lane divided arterials. There are no other consistent results. Table 5 Negative binomial regression models with ADT and lane width for roadway segments in Durres VI. DISCUSSIONS OF RESULTS A. Lane Widths on Arterial Roadway Segments Analysis of geometric design, traffic volume, and accident data collected, has found that, with limited exceptions, there is no consistent, statistically significant relationship between lane width and safety for midblock sections of urban and suburban arterials. There is no indication that the use of 3.0 or 3.3 m, rather than 3.6 m lanes, for arterial midblock segments leads to increases in accident frequency. There are situations in which use of narrower lanes may provide benefits in traffic operations, pedestrian safety, and/or reduced interference with surrounding development, and may provide space for geometric features that enhance safety such as medians or turn lanes. The analysis results indicate narrow lanes can generally be used to obtain these benefits without compromising safety. Two caveats should be noted. First, the data for one of the states analyzed showed an increase in crash rates for four-lane undivided arterials with lane widths of 3.0 m or less, while the data from another state showed an increase in crash rates for four-lane divided arterials with lane widths or 2.7 m, or less. While the results from each state were not confirmed in data from the other state, the findings indicate that lane widths of 3.0 m or less on four- lane undivided arterials and lane widths of 2.7 m or less on four-lane divided arterials should be used cautiously unless local experience indicates otherwise. Second, until more is learned about the interactions between motor vehicles and bicycles on streets with narrower lanes, lane widths less than 3.6 m should be used cautiously, where substantial volumes of bicyclists share the road with motor vehicles, unless an alternative facility for bicycles such as a wider curb lane or paved shoulder is provided. VII. CONCLUSIONS AND RECOMMENDATIONS A safety evaluation of lane widths for arterial roadway segments found no indication, except in limited cases, that the use of narrower lanes increases crash frequencies. The lane width effects in the analyses conducted were generally either not statistically significant. or indicated that narrower lanes were associated with lower rather than higher crash frequencies. There were limited exceptions to this general finding. It was found that crash frequency in one city was higher for 3.0 m lanes than for 3.3 and 3.6 m lanes on fourlane undivided arterials and was higher in the other city for 2.7 m lanes that for 3.0 m lanes on four-lane divided arterials. However, neither of these statistically significant effects observed in one city were statistically significant in the other city. The lane width effects in the three analyses conducted were generally either not statistically significant or inconsistent. With only one limited exception, there is no indication that the use of lanes narrower than 3.6 m on intersection approaches leads to increases in crash frequency. The data for one city showed higher crash frequencies for approaches to four-leg STOP-controlled intersections, for approaches with 3.0 m lanes than for approaches with 3.6 m lanes; however, just the opposite was found in data from the other city. 123
It is concluded from this paper that there is no indication that crash frequencies increase as lane width decreases for arterial roadway segments. These findings suggest that the AASHTO Green Book is correct in providing substantial flexibility for use of lane widths narrower than 3.6 m on urban and suburban arterials. Use of narrower lanes in appropriate locations can provide other benefits to users and the surrounding community including shorter pedestrian crossing distances and space for additional through lanes, auxiliary and turning lanes, bicycle lanes, buffer areas between travel lanes and sidewalks, and placement of roadside hardware. Interpretation of design policies as rigidly requiring the use of 3.6 m lanes on urban and suburban arterials may miss the opportunity for these other benefits without any documentable gain in safety. The paper found three situations in which the observed lane width effect was inconsistent increasing crash frequency with decreasing lane width in one city and the opposite effect in another city. These three situations are: lane widths of 3.0 m or less on four-lane undivided arterials. lane widths of 2.7 m or less on four-lane divided arterials. lane width of 3.0 m or less on approaches to four-leg STOP-controlled arterial intersections. Because of the inconsistent findings mentioned above, it should not be inferred that the use of narrower lane must be avoided in these situations. Rather, it is recommended that narrower lane widths be used cautiously in these situations unless local experience indicates otherwise. REFERENCES [1] Potts, I. B., D.W. Harwood, D. J. Torbic, K. M. Bauer, D. K. Gilmore, D. K. Lay, J. F. Ringert, J. D. Zeeger, D. L. Harkey, and J. M. Barlow, Lane Widths, Channelized Right Turns, and Right-Turn Deceleration Lanes in Urban and Suburban Areas Final Report of NCHRP Project 3-72, Midwest Research Institute, August 2006. [2] Harwood, D. W., F. M. Council, E. Hauer, W. E. Hughes, A. Vogt, Prediction of the Expected Safety Performance of Rural Two-Lane Highways, Report FHWA-RD-99-207, Federal Highway Administration, December 2000. [3] Hauer. E., Safety Models for Urban Four-Lane Undivided Road Segments, presented at the 83rd Annual meeting of the Transportation Research Board, Washington DC, Jannuary 2004. [4] Strathman, J. G., K. J. Dueker, J. Zhang, and T. Williams, Analysis of Design Attributes and Crashes on the Oregon Highway System, Oregon Department of Transportation, Center for Urban Studies, Portland State University, June 2001. [5] Hadi, M. A., J. Aruldhas, L. F. Chow, and J. A. Wattleworth, Estimating Safety Effects of Cross-Section Design for Various Highway Types Using Negative Binomial Regression, Transportation Research Record 1500, Transportation Research Board,1995. [6] American Association of State Highway and Transportation Officials, A Policy on Geometric Design of Highways and Streets, Washington, DC 2001. [7] Harwood, D. W., K. M. Bauer, K. R. Richard, D. K. Gilmore, J. L. Graham, I. B. Potts, D. J. Torbic, and E. Hauer, Methodology to Predict the Safety Performance of Urban and Suburban Arterials, Preliminary Draft Final Report of NCHRP Project 17-26, Midwest Research Institute, April 2006. [8] Zeeger, C. V., D. W. Reinfurt, J. Hummer, L. Herf, and W. Hunter, Safety Effects of Cross-Section Design for Two-Lane Roads, Transportation Research Record 1195, Transportation Research Board, 1988. 124